AN ESTIMATION OF THE ATOMIC CONTRIBUTION TO OCTANOL-WATER PARTITION-COEFFICIENT AND MOLAR REFRACTIVITY FROM FUNDAMENTAL ATOMIC AND STRUCTURAL-PROPERTIES - ITS USES IN COMPUTER-AIDED DRUG DESIGN

It has been shown earlier that atomic physicochemical properties can be extremely valuable in determining the molecular similarities and in modeling the hypothetical ligand-receptor interaction [Ghose et al. (1989) J. Med. Chem. 32 746 and references cited therein]. In all previous studies a discretized atom classification was used to assign the atomic octanol-water partition coefficient and molar refractivity. We have presented here a more continuous empirical method for evaluating these two physicochemical properties using the topological information, and van der Waals radii and electronegativities of the atoms. The method is good for evaluating both the atomic and molecular properties. The basic hypothesis of the method is that these molecular properties are the overall effect of contribution of the individual atoms, although the contributions of the atoms are altered by its environment. We have suggested some functions that can estimate the property of any atom from its environment. Being an atom based approach it is capable of evaluating the property at any local region of a molecule. The contributions for carbon, hydrogen, oxygen, nitrogen, sulfur, phosphorus, selenium and halogens at their various hybridized states have been assigned in terms of 21 atom types. In the case of hydrophobicity (octanol-water partition coefficient) the correction for the neighboring atoms has been modeled in terms of Van der Waals radius, electronegativity and the bonding pattern using 36 adjustable parameters. The observed and the calculated values for a training set of 893 compounds showed a correlation coefficient of 0.909 and an rms deviation of 0.542. The corresponding values for the test set of 127 compounds were 0.817 and 0.620. For molar refractivity the neighboring effect was modeled in terms of van der Waals radius and bonding pattern using 21 adjustable parameters. The training set containing 547 compounds showed a correlation coefficient of 0.997 and an rms deviation of 1.014 between the observed and calculated values. The test set of 62 compounds on the other hand showed the corresponding statistics of 0.995 and 1.605.